GPT-4 Vision Preview: Complete Benchmark Performance Guide

openai 4-vision-preview 2023-09-25 Proprietary

In-depth analysis of GPT-4 Vision Preview performance on MMLU, HumanEval, GSM8K, MATH, BBH benchmarks, including pricing and use cases.

Key Specifications

Vendoropenai
Version4-vision-preview
Release Date2023-09-25
Context Window128000 tokens
Input Modalitiestext, image
Output Modalitiestext
LicenseProprietary
Documentationhttps://platform.openai.com/docs/models

Benchmark Performance

Benchmark Score Unit Evaluated At Notes Source
MMLU 85 % 2023-09-25 5-shot view
HUMANEVAL 70.6 pass@1 2023-09-25 — view
GSM8K 84.5 % 2023-09-25 0-shot CoT view
MATH 49.7 % 2023-09-25 0-shot CoT view
BBH 75 % 2023-09-25 3-shot CoT view
GPQA 36.3 % 2023-09-25 0-shot view
IFEVAL 72.7 % 2023-09-25 prompt_strict view
ARC 94.6 % 2023-09-25 challenge view
MUSR 64.5 % 2023-09-25 0-shot view
WINOGRANDE 80.6 % 2023-09-25 0-shot view

Pricing

Tier Price Currency
Input$10 / MtokUSD
Output$30 / MtokUSD
Cache Read$0 / MtokUSD
Cache Write$0 / MtokUSD

Source: https://openai.com/api/pricing/ · as of 2023-09-25

Compliance

  • Data Residency: US
  • SOC2: ✓
  • HIPAA: ✓
  • GDPR: ✓
  • ISO 27001: ✓

GPT-4 Vision Preview

Model Overview

OpenAI GPT-4 Vision 预览版, 支持图像输入理解, 128K 上下文, 多模态能力首次集成。

Core Specifications

Vendor Version Release Date Context Window Input Modalities Output Modalities License
Openai 4-vision-preview 2023-09-25 128K text, image text Proprietary

Benchmark Performance

Benchmark Score Unit Notes
MMLU (Massive Multitask Language Understanding) 85.0 % 5-shot
HumanEval 70.6 pass@1 —
GSM8K (Grade School Math 8K) 84.5 % 0-shot CoT
MATH 49.7 % 0-shot CoT
BBH (BIG-Bench Hard) 75.0 % 3-shot CoT
GPQA 36.3 % 0-shot
IFEval 72.7 % prompt_strict
ARC 94.6 % challenge
MUSR 64.5 % 0-shot
WinoGrande 80.6 % 0-shot

Pricing

Input Output Cache Read Cache Write
— — — —

per million tokens

Strengths

  • MMLU score 85.0, strong knowledge reasoning.
  • Input Modalities: text, image, audio.

Weaknesses

  • Proprietary, not self-hostable.

Use Cases

  • Code generation and debugging
  • Vision and image understanding
  • Agent workflows and tool use

References